Models Can Prompt Now(simonwillison.net)
simonwillison.net
Models Can Prompt Now
https://simonwillison.net/2025/Sep/14/models-can-prompt/
3 comments
This has been a thing for a while now - particularly around early versions of image generation models. It wasn't entirely uncommon to wire up a small llama-based LLM model into a ComfyUI workflow that would enrich and/or rewrite simpler prompts using SD 1.5 style notation ("hyperrealistic, octane render, 8k, etc").
Makes sense. This is probably one of the few clear examples of models getting better by learning from content online written about themselves.
So definitely it has been improved in this area, intentionally.